This is a list of changes that have been made since the 11/16/88 version

of ksh.

1. New features in 12/28/93

a. Associative arrays. The new version of ksh supports both

associate arrays and the older indexed arrays with the same

array syntax. A new -A option of typeset is used to declare

an array to be associative. As with indexed arrays, $name is

equivalent to ${name[0]}. The prefix operator ! was added

to the parameter expansion syntax to expand to the list of

indices. For example, ${!name[@]} expands to the list of array

indices for variable name.

b. Several additions have been made to shell arithmetic:

1. The shell now performs floating point arithmetic. The

typeset options -F and -E have been added for floating

point and scientific notation respectively.

2. The prefix and postfix ++ and -- operators.

3. The comma and ?: operators.

4. The math library functions.

5. An arithmetic for statement of the form

for ((expr1; expr2; expr3))

do ...

done

6. Integer arithmetic extended up to base 64.

c. Some additions to the macro expansion syntax have been made

to specify substrings and sub-arrays:

1. ${name:expr} expands to the substring of ${name} starting at

the character position defined by arithmetic expression expr.

2. ${name:expr1:expr2} expands to the substring of ${name} starting

at expr1 and consisting of at most expr2 characters.

3. ${name[@]:expr} expands to the values of ${name[@]} starting at

the element defined by arithmetic expression expr.

4. ${name[@]:expr1:expr2} expands to at most expr2 values of

${name} starting at expr1.

5. ${@:expr} expands the positional parameters starting at expr.

6. ${@:expr1:expr2} expands to at most expr2 positional parameters

starting at expr1.

7. ${!name} expands to the name of the variable named by name.

It will expand to name unless name is reference variable.

8. ${!name[sub]} expands to the name of the subscript of the

given variable. If sub is @ or * the list of subscripts

is generated.

9. ${!prefix*} and ${!prefix@} expand to the list of variable

names beginning with prefix.

10. The substring operators, # and % can be now be applied

with aggregates (@ or *) and are applied to each.

11. ${name/pattern/string} expands to the value of name with

the first occurrence of pattern replaced by string.

With aggregates (@ or *) this operation is applied to each.

12. ${name/#pattern/string} Same as above but the pattern

to be replaced must match at the beginning.

13. ${name/%pattern/string} Same as above but the pattern

to be replaced must match at the end.

14. ${name//pattern/string} expands to the value of name with

the each occurrence of pattern replaced by string.

With aggregates (@ or *) this operation is applied to each.

d. The name space for variables has been extended. The character '.'

can be used at the beginning of a name, and to separate identifiers

within a name. However, to create a name of the form, foo.bar,

the variable foo must exist. The namespace starting with .sh

is reserved for shell implementation variables. Exported

variable cannot contain a '.'.

e. Compound assignments. The assignment syntax, varname=value,

has been extended to allow assignments of the form

varname=(assignment_list). As elsewhere in the shell

spaces or tabs are optional around the parentheses, and

no space is permitted between the varname and the =. The

assignment_list can be one of the following:

1. A list of words. In this case each word is expanded as

in a for list and the resulting items become elements

of the indexed array varname.

2. A list of subscript assignments in the form

[subscript]=value. In this, these elements become

elements of the associative array varname.

3. A list of assignments; simple or compound. In this

case, each assignment is made to varname.name, where

name is the name of the enclosed assignment.

4. Assignments in the form of readonly or typeset

statements. In this case each assignment is made as

in 3 above, and the attributes are given to the

corresponding variable.

In case 3 and 4 above, the value of "$varname" after

the above assignment is (assignment_list), where the

assignment_list produced would reproduce all of the

variables under varname.*.

f. Function names of the form variable.action (called discipline

functions) can be defined where variable is any valid variable

name and action is get, set, or unset. The function variable.get

is invoked each time the variable is referenced. The set

discipline is invoked each time the variable is assigned to.

The unset discipline is invoked when a variable is unset.

The new variables .sh.name, .sh.subscript, and .sh.value are

defined inside the function body. Other shell extensions

may have their own set of discipline functions.

g. The compound command !, which negates the return value of the

following pipeline, has been added.

h. On systems that support dynamic loading with dlopen(), it is

now possible to add built-in commands at runtime with the

a builtin command named builtin.

i. The following builtins have been added:

1. command name [ ... ]

2. sleep [decimal-seconds]

3. builtin [-ds] [-f file] [name...]

4. getconf name [pathname]

5. disown [job...]

j. An addition format for literal strings, $'....' can

be used where ever literal strings are valid. The string

inside the single quotes will be converted using the ANSI-C

escape conventions. Additionally, the escape sequence \E

expands to the escape character (default \033) whenever ANSI-C

escape sequences are recognized.

k. A typeset -n option has been added which causes the value of a

variable to be treated as a reference to another variable so that

variables can be indirectly named. For example, if $1 contains

the name of a variable, then typeset -n foo=$1 causes the variable

foo to be synonymous with the variable whose name is $1. A builtin

alias, nameref='typeset -n' has been added to aid mnemonics.

Reference names cannot contain a '.'. Whenever that portion of

a variable up to the first '.' matches a reference name, the

reference value is substituted. For example, with nameref foo=.top,

then ${foo.bar} is equivalent to ${.top.bar}. When used as the

index of a for or select loop, each assignment causes a

new name reference to occur.

l. The KEYBD trap has been added which is triggered when a key

or escape sequence is typed while reading from the keyboard

in an edit mode. This, combined with some new variables

makes it possible to program your key bindings in ksh.

m. New variables have been added:

1. FIGNORE defines a set of file names to be ignored in each

directory when performing pathname expansion, replacing

the rule that requires that a leading . be matched explicitly.

2. Variable sh.edchar contains the value of the keyboard character

that has been entered when processing a KEYBD trap. If the value

is changed as part of the trap action, then the new value replaces

the key or keys that caused the trap.

3. Variable sh.edcol is set to the character position of the cursor

within the input buffer during a KEYBD trap.

4. Variable sh.edmode is set to the escape character when in vi

insert mode.

5. Variable sh.edtext is set to the contents of the input buffer

during a KEYBD trap.

6. HISTEDIT is checked before FCEDIT. FCEDIT is obsolete.

7. HISTCMD is the number of the current command in the history

file.

8. Variable .sh.version is set to the version string for

this shell.

9. Variable .sh.name is set to the name of the variable

that that was referenced or assigned to when executing a get

or set discipline function.

10. Variable .sh.subscript is set to the subscript for the variable

that was referenced or assign to when executing a get or

set discipline function.

11. Variable .sh.value is set to the new value for the variable

that was assigned to when executing the set discipline function.

n. New invocation and set -o options have been added:

1. set -o notify (or set -b) causes background completion messages

to be displayed as soon as the job completes.

2. There is a compile time option named KIA which enables

creation of a relational database for commands, variables

and functions defined and referenced by a script. The

option -I <filename>, causes the database to be generated

in <filename>. The database format can be queried via

the cql command.

o. ksh93 can read and evaluate pre-compiled scripts generated by

a separate program called shcomp.

p. More work on internationalization has been added:

1. The decimal point character is processed per locale

2. A $ can be placed in front of each string to indicate

that the string needs translation but is otherwise ignored.

This means that if a message catalog of all $"..." strings

is generated, then a program such as print $"hello world"

could display "bonjour monde" in the french locale.

q. Backreferences have been added to pattern matching. The sequence

\d, where d is a digit from 1-9, matches the same string as

the d-th previous parenthesis group. Backreferences

can be used within patterns, and within replacement strings

with any of the ${name/...} operators.

2. Changes made in 12/28/93

a. The output format of many commands has changed as follows:

1. System error messages are displayed whenever a failure

is caused by a system call.

2. The exit status has changed in many cases:

a. USAGE messages cause an exit status of 2.

b. Commands not found cause exit - 127.

c. Command found, but not executable - 126.

d. Terminated because of signal - 256+sig

3. The output of values from built-ins that contain special

characters are quoted in a manner that then can be re-input.

4. The trace output puts quotes around the output so that it

can be reused as input.

5. The output for trap is in a format that can be reinput the

the shell to restore the traps.

6. kill -l lists the signal names without numbers as

required by the POSIX standard.

b. The following changes have been made to shell functions:

1. The semantics of functions declared with name() has changed

to conform with the IEEE-POSIX 1003.2 standard. In particular,

these functions are executed in a dot script environment rather

than a separated function environment so that there are no

local variables and no scoping for traps.

2. Functions declared as function name, preserve the old ksh

semantics can be also used as the first argument to the dot (.)

command to have them executed in a dot script environment.

c. The command search rules have changed as follows:

1. Special built-ins (those with a dagger in front of them) are

executed first.

2. Functions are executed next.

3. Other built-ins that do not require an executable version

(for example cd and read) come next.

4. If the command name contains a slash, the pathname corresponding

to the command name is executed.

5. If name corresponds to a previously encountered pathname

on the PATH variable, the corresponding command is executed.

6. If the command name does not contain a slash, then the PATH

variable is used to find an executable by that name. If

the directory that the command is found is also contained in

the FPATH variable, then the command treated as a function.

If the shell has a built-in version of the command corresponding

to this command, then the built-in version of this command

is executed. Otherwise, the shell remembers that pathname

corresponding to this command name and executes this pathname.

7. If the name is not found on PATH, then the directories in

FPATH are searched. If found, then the command is executed

as a function.

d. Built-in commands options now conform to the IEEE-POSIX 1003.2

conventions with some additions. In particular,

name -?

will now print a Usage line for name, except for true, false,

colon, login, newgrp, echo, [, and command.

e. Tilde expansion is now performed as part of the word expansions.

The effect of this is that if word begins with ~ in ${name op word},

it will be expanded unless escaped.

f. Pathname expansion is no longer performed on redirection words

unless the shell is interactive.

g. Changes to shell and options:

1. The -n option has been enhanced to produce more warning and

portability messages.

2. The -C option is equivalent to -o noclobber. Files are

created with O_EXCL when -C is on.

h. The following changes have been made to [[...]]:

1. A string by itself is equivalent to -n string.

2. -e has been added as equivalent to -a.

3. == has been added as equivalent =.

4. -a and = are now considered obsolete.

5. Arithmetic comparisons are now considered obsolete.

i. kill has been changed as follows:

1. Signal names can be upper case or lower case.

2. Numerical arguments to kill -l cause the given signal names to

be displayed.

3. String arguments to kill -l cause the given signal numbers to

be displayed.

4. Synopsis changed for getopts conformance.

j. print has a -f format option which is equivalent to

the IEEE POSIX printf. Both print -f format, and

printf have the following extensions from IEEE POSIX:

1. Floating point formats are supported.

2. Size and precision specifications can be *.

3. The %d option can take an argument after precision to

specify the base that the number will be displayed.

4. A %q format can be used to output a string quoted so

that it can be re-input to the shell.

5. A %P format can be used to output the shell pattern which

corresponds to the give extended regular expression.

6. For numerical fields, the arguments can be arithmetic

expressions which will be evaluated.

7. The %n format works as described in ANSI-C.

k. The following changes have been made to fc:

1. It has been renamed hist. fc is now a predefined alias.

2. hist uses ${HISTEDIT:-$FCEDIT}. FCEDIT is obsolete.

3. A new -s option is equivalent to the obsolete -e -.

4. If the first argument refers to a command earlier than the

first accessible command, it now implies the first accessible

command, so that hist -l 1 lists all accessible history commands.

l. The dot command (.) has changed as follows:

1. The argument can be the name of a function declared as

function name. The function will execute without creating a

new scope.

2. If there are arguments to the given script or function,

the positional parameters are restored to their original

value when . completes.

m. The read built-in has been changed as follows:

1. A -A option to read has been added to allow the fields to be

read into an indexed array.

2. A -t n option has been added which causes read to

timeout after n seconds when reading from a slow device.

3. A -d char option has been added which causes the read

to terminate at char rather than at new-line.

n. The trap command has been changed as follows:

1. Trap names can be either upper case or lower case.

2. Trap -p cause only the specified trap values to be displayed.

3. The value of trap in a subshell will be the value in the parent

shell until a call to trap which changes the trap settings has

been made. Thus, savetraps=$(trap) works as required by the

POSIX standard.

o. The exec command has been extended as follows:

1. The -c option clears the environment first.

2. The -a name option sets argv[0] to name for the program.

p. true and false are built-ins, not aliases to built-ins.

q. test has been modified to conform to the IEEE-POSIX 1003.2

standard when there are three or less arguments.

r. umask -S option displays the mask in a symbolic format.

s. wait now returns the correct exit status of any previous

background job that has not been waited for, not just

the most recent one.

t. The whence built-in has an option -a which causes all

uses for the given command name to be reported.

u. unalias has -a option to clear all the aliases.

v. The times built-in command has been removed. The time

reserved word, without a command, gives time cumulative

time for the shell and its children. A built-in alias

for times should enable scripts using times to continue

to run.

w. Command substitution and arithmetic substitution will now be

performed for PS1, ENV, and PS4 evaluation in addition to

parameter expansion.

x. The SECONDS variable now displays elapsed time in floating

point seconds with 3 places after the decimal point by

default.

y. The getopts built-in now handles the complete libast optget

functionality. If any errors have occurred with getopts

when it has reached the end of arguments, then the Usage

message will be generated from the option string and the

exit status from getopts will be 2 rather than 1. The

usage message will be stored in the OPTARG variable if

the option string contains a leading colon; otherwise

it will be printed on standard error automatically.

z. THE ENV file is only processed for interactive shell

invocations. In addition, the -x attributes for

aliases and functions is ignored.

aa. The built-in edit modes have been changed as follows:

1. The pathname completion and pathname listing options

now perform command completion and command listing

when applied to a word in the command position.

2. In emacs mode ^N as the first related command after

the prompt will move to the next command relative to the

last known history position.

3. In emacs mode, successive kill and delete commands will

accumulate their data in the kill buffer, by appending or

prepending as appropriate. This mode will be reset by any

command not adding something to the kill buffer.

4. The control-T of emacs mode has been changed to behave like

control-T in gnu-emacs.

bb. The TMOUT variable also sets a limit for select timeouts

and default timeouts for read.

4. The source code has undergone significant modification.

a. Much of the code has been rewritten, In many cases this has

resulted in significant performance improvement.

b. The code is organized differently. See the README files

for more details.

c. Most configuration parameters now get generated using

the FEATURE mechanism of nmake. Other options are set

in the OPTIONS file.

c. The are several new compile time options. See the README

file for details. Some of the old ones have been removed.

d. The install script is a Mamfile that is generated by

nmake and processed by a script that comes with the

distribution.

e. There are far fewer global names. This should make it

must easier to add built-in commands without worrying

about conflicts.

f. The code uses the sfio library which makes it possible

to mix with stdio.

g. The code is written in ANSI C with full prototypes.

The code is based on the IEEE POSIX 1003.1 standard.

The code can be compiled with K&R C and with C++ by

using the ANSI cpp that comes with nmake or running

the code through the proto filter before pre-processing.

This happens automatically with our shipping system.

h. There is a programming interface for capturing references

and assignment to shell variables. It is also possible

to intercept variable creation and supply the array processing

function for that variable. See nval.3 for a description.